Continue.dev | Northern Inference docs

Get going in 30 seconds

Open ~/.continue/config.yaml and add:

models:
  - name: Claude Haiku 4.5 (NI)
    provider: openai
    model: bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0-global
    apiBase: https://northerninference.ca/v1
    apiKey: ni_live_YOUR_KEY_HERE

Reload VS Code. Pick Claude Haiku 4.5 (NI) from the Continue model picker and send a message. Check portal → Usage to confirm the request landed on NI.

Continue is an open-source VS Code / JetBrains plugin. It speaks OpenAI-compatible out of the box.

Config

Open ~/.continue/config.yaml (or use Continue → Settings → Open Config):

name: Northern Inference
version: 0.0.1

models:
  - name: Gemini 2.5 Pro (NI, CA route)
    provider: openai
    model: vertex_ai/gemini-2.5-pro-ca
    apiBase: https://northerninference.ca/v1
    apiKey: ni_live_YOUR_KEY_HERE

  - name: Claude Haiku 4.5 (NI, global route)
    provider: openai
    model: bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0-global
    apiBase: https://northerninference.ca/v1
    apiKey: ni_live_YOUR_KEY_HERE

  - name: Claude Sonnet 4.6 (NI, global route)
    provider: openai
    model: bedrock/global.anthropic.claude-sonnet-4-6-global
    apiBase: https://northerninference.ca/v1
    apiKey: ni_live_YOUR_KEY_HERE

tabAutocompleteModel:
  name: Claude Haiku 4.5 (autocomplete)
  provider: openai
  model: bedrock/global.anthropic.claude-haiku-4-5-20251001-v1:0-global
  apiBase: https://northerninference.ca/v1
  apiKey: ni_live_YOUR_KEY_HERE

Key points

**provider: openai**: Continue treats any OpenAI-compatible endpoint this way.
Model IDs are route keys. Copy the exact key from /models. API key allowed route tiers authorize which routes the key may call.
**tabAutocompleteModel**: use a cheap, fast model (Haiku) for this. Every keystroke hits the API.

Verify

In Continue's chat panel, ask:

Which model are you?

Then check portal → Usage for the request to appear.

Cost control

Continue fires a request on almost every keystroke when autocomplete is enabled. A separate low-budget key for the autocomplete model is a good idea: portal → Keys → Create key → Daily spend cap.